cgn-ctl

Admin CLI

Cluster bring-up, model deploys, traffic shaping, and debugging on bare metal. Optional Helm when you deploy on Kubernetes.

Overview

cgn-ctl is the operational command-line interface for Cognitora Inference. It drives bare-metal and systemd installs first, provides direct etcd access for routing policy management, and offers cluster introspection commands. Optional Helm and CRD reads keep Kubernetes workflows aligned with the same policies — no hidden state between CLI and declarative rollouts.

Features

Bare-metal cluster bring-up and systemd-oriented workflows
Optional Helm for Kubernetes cluster management
Writes routing policies directly to etcd
cgn-ctl cluster set-policy for live weight tuning
Model deployment and lifecycle management
Traffic shaping and split configuration
Cluster health introspection and debugging
Reads the same CRDs as cgn-operator when Kubernetes is in use
Version pinning and upgrade management

Architecture

cgn-ctl → etcd (policy writes, cluster state reads) | cgn-ctl → bare-metal / systemd targets | optional cgn-ctl → Helm + Kubernetes API (CRD reads). Single static binary with no runtime dependencies.

Configuration

Key	Type	Default	Description
cluster.etcd_endpoints	array	["http://127.0.0.1:2379"]	etcd cluster endpoints
cluster.name	string	""	Target cluster name

Example

bash

# cgn-ctl uses CLI flags or a config file
cgn-ctl --cluster production cluster status
cgn-ctl cluster set-policy --kv 0.6 --load 0.2 --power 0.1 --capacity 0.1
cgn-ctl deploy llama-3.1-70b --replicas 4 --engine vllm --tp 4
cgn-ctl install --namespace cognitora-system

cgn-metrics

cgn-operator